AI, agents and agentic glossary
This page is for readers who meet an unfamiliar term in this section. The general glossary of Exponential is Glossary.
Contents
A to F
- Agent: a program in which an AI model decides which action to take next, usually by calling tools. Agentic workflows.
- Agentic workflow: a task carried out by an agent over several steps, such as a report or an import. A content import workflow.
- Audit trail: the kernel's tamper-evident record of who did what; the assistant writes to the channel
ai. The audit trail for AI. - Chain (AI chain): multi-step writing through the job queue with a human checkpoint. Chains and queued work.
- Data policy: rules that decide which content may go to which provider, by region and retention. Privacy and data leaving the site.
- Draft: an unpublished version; everything Exponential Intelligence writes is one. Drafts and human review.
- EI: Summary: the printable page for a data protection officer (
expai/summary). - expai: the extension name of Exponential Intelligence. Exponential Intelligence for AI users.
- expservices: the extension that offers administration functions as named services. Agents and expservices.
- Exponential Intelligence: the writing assistant of the administration interface.
- Fallback: when a chosen model cannot answer, the site default answers instead. Providers and models.
G to O
- Label: the mark "Exponential Intelligence" on every version the assistant wrote.
- Lane: a way of sending a request (standard, flex, batch, fast) with its own price and speed. Cost control.
- Local model: an open-weight model running on your own server; region
local. Local models. - Model: the AI system that writes text; reached through a provider.
- OAuth2 token: a REST credential for an application that acts for users who sign in themselves.
P to Z
- Personal API key: a REST credential
expk_...with scopes, acting as its owner. A read-only API key, step by step. - Personal token (expservices): a credential
expt_...for expservices, acting as its user. - Plan mode: Vector's mode in which the model proposes a plan with a cost estimate before bulk work. Vector and staged review.
- Probe: 15 scripted tasks a model must pass before it may write drafts. The write gate and the probe.
- Prompt injection: text in content written to steer a model. Untrusted data and prompt injection.
- Provider: the service (or local server) that runs a model. Providers and models.
- Region / retention: where a provider processes requests (
us,eu,local) and how long it keeps them (none,zdr,30d,unknown). - Scope (API key): what a key may be used for (
read,publish,edit,remove), each standing for a policy of its owner. - Scope (Vector): the nodes, subtrees, classes and languages a thread may touch.
- Service tool: a read service of expservices that the assistant may call. Service tools.
- Staged review: proposals reviewed per field and part before they are written into drafts.
- Token (billing): the unit providers count and bill. Token (credential): a secret that authenticates.
- Tool: one named ability a model can call, with a description and an input schema.
- Vector: the workspace of threads of Exponential Intelligence for larger jobs.
- Veto hook: rules or a listener that refuse single AI requests with a reason. The master switch and the veto hook.
- Web tools:
web_fetch,web_searchandproject_historyof the assistant. Web and history tools. - Write gate: the rule that decides which models may write drafts.
Related pages
- AI, agents and agentic: FAQ
- Glossary
- Free documentation: this page is free to copy, translate and improve under the GNU FDL 1.3.
Become an Exponential Partner
Partners share their information, provide funding, contribute back and develop the products with us, and that is how Exponential grows. A partner website and dashboard are planned for mid-2027: please become a partner today to speed up our growth rate and land new features faster.